Automatic Root Length Estimation from Images Acquired In Situ without Segmentation

Image-based root phenotyping technologies, including the minirhizotron (MR), have expanded our understanding of the in situ root responses to changing environmental conditions. The conventional manual methods used to analyze MR images are time-consuming, limiting their implementation. This study presents an adaptation of our previously developed convolutional neural network-based models to estimate the total (cumulative) root length (TRL) per MR image without requiring segmentation. Training data were derived from manual annotations in Rootfly, commonly used software for MR image analysis. We compared TRL estimation with 2 models, a regression-based model and a detection-based model that detects the annotated points along the roots. Notably, the detection-based model can assist in examining human annotations by providing a visual inspection of roots in MR images. The models were trained and tested with 4,015 images acquired using 2 MR system types (manual and automated) and from 4 crop species (corn, pepper, melon, and tomato) grown under various abiotic stresses. These datasets are made publicly available as part of this publication. The coefficients of determination (R2), between the measurements made using Rootfly and the suggested TRL estimation models were 0.929 to 0.986 for the main datasets, demonstrating that this tool is accurate and robust. Additional analyses were conducted to examine the effects of (a) the data acquisition system and thus the image quality on the models' performance, (b) automated differentiation between images with and without roots, and (c) the use of the transfer learning technique. These approaches can support precision agriculture by providing real-time root growth information.

Semantic segmentation of plant roots from RGB (mini-) rhizotron images-generalisation potential and false positives of established methods and advanced deep-learning models

BACKGROUND

Manual analysis of (mini-)rhizotron (MR) images is tedious. Several methods have been proposed for semantic root segmentation based on homogeneous, single-source MR datasets. Recent advances in deep learning (DL) have enabled automated feature extraction, but comparisons of segmentation accuracy, false positives and transferability are virtually lacking. Here we compare six state-of-the-art methods and propose two improved DL models for semantic root segmentation using a large MR dataset with and without augmented data. We determine the performance of the methods on a homogeneous maize dataset, and a mixed dataset of > 8 species (mixtures), 6 soil types and 4 imaging systems. The generalisation potential of the derived DL models is determined on a distinct, unseen dataset.

RESULTS

The best performance was achieved by the U-Net models; the more complex the encoder the better the accuracy and generalisation of the model. The heterogeneous mixed MR dataset was a particularly challenging for the non-U-Net techniques. Data augmentation enhanced model performance. We demonstrated the improved performance of deep meta-architectures and feature extractors, and a reduction in the number of false positives.

CONCLUSIONS

Although correction factors are still required to match human labelled root lengths, neural network architectures greatly reduce the time required to compute the root length. The more complex architectures illustrate how future improvements in root segmentation within MR images can be achieved, particularly reaching higher segmentation accuracies and model generalisation when analysing real-world datasets with artefacts-limiting the need for model retraining.

Evaluation of functional tests performance using a camera-based and machine learning approach

The objective of this study is to evaluate the performance of functional tests using a camera-based system and machine learning techniques. Specifically, we investigate whether OpenPose and any standard camera can be used to assess the quality of the Single Leg Squat Test and Step Down Test functional tests. We recorded these exercises performed by forty-six healthy subjects, extract motion data, and classify them to expert assessments by three independent physiotherapists using 15 binary parameters. We calculated ranges of movement in Keypoint-pair orientations, joint angles, and relative distances of the monitored segments and used machine learning algorithms to predict the physiotherapists' assessments. Our results show that the AdaBoost classifier achieved a specificity of 0.8, a sensitivity of 0.68, and an accuracy of 0.7. Our findings suggest that a camera-based system combined with machine learning algorithms can be a simple and inexpensive tool to assess the performance quality of functional tests.

Using EfficientNet-B7 (CNN), Variational Auto Encoder (VAE) and Siamese Twins' Networks to Evaluate Human Exercises as Super Objects in a TSSCI Images

In this article, we introduce a new approach to human movement by defining the movement as a static super object represented by a single two-dimensional image. The described method is applicable in remote healthcare applications, such as physiotherapeutic exercises. It allows researchers to label and describe the entire exercise as a standalone object, isolated from the reference video. This approach allows us to perform various tasks, including detecting similar movements in a video, measuring and comparing movements, generating new similar movements, and defining choreography by controlling specific parameters in the human body skeleton. As a result of the presented approach, we can eliminate the need to label images manually, disregard the problem of finding the start and the end of an exercise, overcome synchronization issues between movements, and perform any deep learning network-based operation that processes super objects in images in general. As part of this article, we will demonstrate two application use cases: one illustrates how to verify and score a fitness exercise. In contrast, the other illustrates how to generate similar movements in the human skeleton space by addressing the challenge of supplying sufficient training data for deep learning applications (DL). A variational auto encoder (VAE) simulator and an EfficientNet-B7 classifier architecture embedded within a Siamese twin neural network are presented in this paper in order to demonstrate the two use cases. These use cases demonstrate the versatility of our innovative concept in measuring, categorizing, inferring human behavior, and generating gestures for other researchers.

Assessing Human Mobility by Constructing a Skeletal Database and Augmenting it Using a Generative Adversarial Network (GAN) Simulator

This paper presents a neural network simulator based on anonymized patient motions that measures, categorizes, and infers human gestures based on a library of anonymized patient motions. There is a need for a sufficient training set for deep learning applications (DL). Our proposal is to extend a database that includes a limited number of videos of human physiotherapy activities with synthetic data. As a result of our posture generator, we are able to generate skeletal vectors that depict human movement. A human skeletal model is generated by using OpenPose (OP) from multiple-person videos and photographs. In every video frame, OP represents each human skeletal position as a vector in Euclidean space. The GAN is used to generate new samples and control the parameters of the motion. The joints in our skeletal model have been restructured to emphasize their linkages using depth-first search (DFS), a method for searching tree structures. Additionally, this work explores solutions to common problems associated with the acquisition of human gesture data, such as synchronizing activities and linking them to time and space. A new simulator is proposed that generates a sequence of virtual coordinated human movements based upon a script.

Parametric temporal compression of infrared imagery sequences containing a slow-moving point target

Infrared (IR) imagery sequences are commonly used for detecting moving targets in the presence of evolving cloud clutter or background noise. This research focuses on slow-moving point targets that are less than one pixel in size, such as aircraft at long range from a sensor. Since transmitting IR imagery sequences to a base unit or storing them consumes considerable time and resources, a compression method that maintains the point target detection capabilities is highly desirable. In this work, we introduce a new parametric temporal compression that incorporates Gaussian fit and polynomial fit. We then proceed to spatial compression by spatially applying the lowest possible number of bits for representing each parameter over the parameters extracted by temporal compression, which is followed by bit encoding to achieve an end-to-end compression process of the sequence for data storage and transmission. We evaluate the proposed compression method using the variance estimation ratio score (VERS), which is a signal-to-noise ratio (SNR)-based measure for point target detection that scores each pixel and yields an SNR scores image. A high pixel score indicates that a target is suspected to traverse the pixel. From this score image we calculate the movie scores, which are found to be close to those of the original sequences. Furthermore, we present a new algorithm for automatic detection of the target tracks. This algorithm extracts the target location from the SNR scores image, which is acquired during the evaluation process, using Hough transform. This algorithm yields a similar detection probability (PD) and false alarm probability (PFA) of the compressed sequences and the original sequences. The parameters of the new parametric temporal compression successfully differentiate the targets from the background, yielding high PDs (above 83%) with low PFAs (below 0.043%) without the need to calculate pixel scores or to apply automatic detection of the target tracks.

Compression of infrared imagery sequences containing a slow-moving point target, part II

Infrared (IR) imagery sequences are commonly used for detecting moving targets in the presence of evolving cloud clutter or background noise. This research concentrates on slow-moving point targets that are less than one pixel in size, such as aircraft at long ranges from a sensor. Because transmitting IR imagery sequences to a base unit or storing them consumes considerable time and resources, a compression method that maintains the point-target detection capabilities is highly desirable. In our previous work, we introduced two temporal compression methods that preserve the temporal profile properties of the point target in the form of discrete cosine transform (DCT) quantization and parabola fit. In the present work, we extend the compression task method of DCT quantization by applying spatial compression over the temporally compressed coefficients, which is followed by bit encoding. We evaluate the proposed compression method using a signal-to-noise ratio (SNR)-based measure for point target detection and find that it yields better results than the compression standard H.264. Furthermore, we introduce an automatic detection algorithm that extracts the target location from the SNR scores image, which is acquired during the evaluation process and has a probability of detection and a probability of false alarm close to those of the original sequences. We previously determined that it is necessary to establish a minimal noise level in the SNR-based measure to compensate for smoothing that is induced by the compression. Here, the noise level calculation process is modified in order to allow detection of targets traversing all background types.

Compression of infrared imagery sequences containing a slow-moving point target

Infrared imagery sequences are used for detecting moving targets in the presence of evolving cloud clutter or background noise. This research concentrates on slow-moving point targets that are less than one pixel in size, such as aircraft at long ranges from a sensor. The infrared (IR) imagery sequences that are captured by ground sensors contain an enormous amount of data. Since transmitting this data to a base unit or storing it consumes considerable time and resources, a compression method that maintains the point target detection capabilities is desired. For this purpose, we developed two temporal compression methods that preserve the temporal profile properties of the point target. We evaluated the proposed compression methods using a signal-to-noise-ratio (SNR)-based measure for point target detection and showed that the compression may improve the SNR results compared to the IR sequence prior to compression.

Observer variability and applicability of BI-RADS terminology for breast MR imaging: invasive carcinomas as focal masses

OBJECTIVE

The purpose of this study was to assess whether the descriptive terminology and final assessment categories of the Breast Imaging Reporting and Data System (BI-RADS) lexicon can be used for breast carcinomas detected on MR imaging and to assess the inter- and intraobserver variabilities in the use of the descriptors and final assessment categories.

MATERIALS AND METHODS

In 82 patients, 101 masses, including 68 infiltrating carcinomas and 33 benign lesions, were interpreted independently by four radiologists and described by BI-RADS terminology with respect to mass shape and margin and BI-RADS final assessment categories. The enhancement pattern of the mass was also reported. In addition, two radiologists interpreted each case twice to evaluate intraobserver variability. The final case set for analysis was the 68 infiltrating carcinomas.

RESULTS

Most of the infiltrating carcinomas were described as irregular, spiculated, and heterogeneously enhancing masses. The final impression of the 68 carcinomas was BI-RADS category 5 (highly suggestive of malignancy) in 41 (61%), category 4 (suspicious abnormality) in 24 (35%), and category 3 (probably benign) in three (4%). Enhancement pattern was heterogeneous in 40 (59%), homogeneous in 14 (21%), and rim in 14 (21%). Interobserver agreement was moderate for mass margin, shape, enhancement, and final assessment category.

CONCLUSION

This study suggests that the mammographic BI-RADS lexicon with some modifications may be applied to describe the features of infiltrating carcinoma seen on breast MR imaging.

Testicular microlithiasis: what is its association with testicular cancer?

PURPOSE

To determine the prevalence of testicular microlithiasis in patients who were referred for scrotal ultrasonography (US) at a tertiary care cancer center and to evaluate the association between microlithiasis and cancer.

MATERIALS AND METHODS

Testicular sonograms obtained in 528 men were retrospectively reviewed to identify patients with US findings suggestive of microlithiasis, intratesticular masses, and intratesticular heterogeneous changes. The association of US findings with medical records and with histopathologic findings that were available in 95 patients was evaluated. Statistical analysis was performed to determine the relationship of testicular cancer, intratesticular mass, and microlithiasis.

RESULTS

Forty-eight (9%) of the 528 patients had microlithiasis; 13 of these (27%) had testicular cancers. Of the 480 patients without microlithiasis, 38 (8%) had testicular cancer. Ninety patients had an intratesticular mass, of whom 23 (26%) had microlithiasis. Forty-three (12 with microlithiasis) patients with a mass had testicular cancer, 43 (10 with microlithiasis) had benign findings or nontesticular malignant histopathologic findings, and four (one with microlithiasis) had no pathologic findings.

CONCLUSION

Intratesticular microlithiasis is highly associated with confirmed testicular cancer, as well as with US evidence of testicular mass.

Medical image restoration of dynamic lungs using optical transfer function of lung motion

When carrying out medical imaging based on detection of isotopic radiation levels of internal organs such a lungs or heart, distortions, and blur arise as a result of the organ motion during breathing and blood supply. Consequently, image quality declines, despite the use of expensive high resolution devices and, such devices are not exploited fully. A method with which to overcome the problem is image restoration. Previously, we suggested and developed a method for calculating numerically the optical transfer function (OTF) for any type of image motion. The purpose of this research is restoration of original isotope images (of the lungs) by restoration methods that depend on the OTF of the real time relative motion between the object and the imaging system. This research uses different algorithms for the restoration of an image, according to the OTF of the lung motion, which is in several directions simultaneously. One way of handling the three-dimensional movement is to decompose the image into several portions, to restore each portion according to its motion characteristics, and then to combine all the image portions back into a single image. An additional complication is that the image was recorded at different angles. The application of this research is in medical systems requiring high resolution imaging. The main advantage of this approach is its low cost versus conventional approaches.

Metastatic lobular carcinoma of the breast: patterns of spread in the chest, abdomen, and pelvis on CT

OBJECTIVE

We determined the pattern of spread of metastatic lobular carcinoma in the chest, abdomen, and pelvis on CT.

MATERIALS AND METHODS

We identified 57 women (age range, 30-79 years; mean age, 57 years) with metastatic lobular carcinoma of the breast who underwent CT of the chest, abdomen, or pelvis between 1995 and 1998. Then two experienced oncology radiologists retrospectively reviewed 78 CT examinations of those patients to identify sites of metastatic disease and to identify complications caused by metastases.

RESULTS

Metastases were identified in bone in 46 patients (81%), lymph nodes in 27 patients (47%), lung in 19 patients (33%), liver in 18 patients (32%), peritoneum in 17 patients (30%), colon in 15 patients (26%), pleura in 13 patients (23%), adnexa in 12 patients (21%), stomach in nine patients (16%), retroperitoneum in nine patients (16%), and small bowel in six patients (11%). Eighteen patients (32%) had gastrointestinal tract involvement that manifested as bowel wall thickening. Hydronephrosis was present in six patients (11%).

CONCLUSION

Although lobular carcinoma metastasized to common metastatic sites of infiltrating ductal carcinoma, lobular carcinoma frequently metastasized to unusual sites, including the gastrointestinal tract, peritoneum, and adnexa. Gastrointestinal tract involvement was as frequent as liver involvement, appearing as bowel wall thickening on CT. Hydronephrosis was a complication of metastatic lobular carcinoma.

Angular dependence of sampling modulation transfer function

Sampling modulation transfer function (MTF) as defined in Park et al. [Appl. Opt. 23, 2527-2582 (1984)] as an x and y sampling can be generalized for image data not along x and y directions. For a given sampling lattice (such as in a laser printer, a scene projector, or a focal-plane array), we construct a two-dimensional sampling MTF based on the distance between nearest samples in each direction. Because the intersample distance depends on direction, the sampling MTF will be best in the directions of highest spatial sampling and poorer in the directions of sparse sampling. We compare hexagonal and rectangular lattices in terms of their equivalent spatial frequency bandwidth. We filter images as a demonstration of the angular-dependent two-dimensional sampling MTF.